Fig. 3: Transitions between thoughts.

a Sentence-level transition probability between different thought categories. The rows and columns of the matrix represent the current and next categories, respectively. The numbers in the matrix indicate transition probabilities for each category pair, averaged across participants. The colormap of the matrix indicates the t-statistics from two-tailed paired t-tests against the chance probability (i.e., the overall proportion of the next category among all sentences within each participant). Transitions that occur more frequently than chance are shown in red, while those that occur less frequently than chance are shown in blue. Full statistics for individual cells, including exact p values, are reported in Supplementary Table 9. *p < 0.05 (Bonferroni corrected). b Measuring semantic similarity between thought units. Each thought unit was converted to a text embedding vector using the Sentence Transformers Python module (version 2.2.0). Semantic similarity between thoughts was defined as the cosine similarity between their embedding vectors. c Semantic similarity as a function of the temporal distance from a target thought unit in each thought category. Lags are measured in units of thought, with lag = 0 representing the target thought. Negative and positive lags indicate thoughts that occurred before and after the target thought, respectively. Solid lines indicate the mean across participants (N = 98, 117, 113, 113, and 112 for current, semantic-world, semantic-self, episodic, and future categories, respectively). Shaded areas indicate the SEM across participants. d Measuring thought boundary agreement scores from think-aloud transcripts. Independent coders assigned the same numbers to consecutive sentences/clauses describing a single thought. Thought boundaries (red bars) were detected when the thought identification numbers changed. Boundary agreement scores were defined as the proportion of coders who identified each moment as a thought boundary. e Mean boundary agreement scores for different types of thought transitions. f Mean semantic similarity between pre- and post-boundary thoughts for different types of thought transitions. In both e and f, each colored dot represents an individual participant (N = 117, 118, and 117 for category change, topic change, and both change conditions, respectively). Black circles indicate the mean across participants within each transition type. Error bars show the SEM across participants. Statistical significance reflects differences between thought transition types, as determined by two-tailed paired t-tests. ***p < 0.001 (uncorrected).